Stray Thoughts #2

Deepfakes FTW; New Battle Reports; Recommendations

Greetings, all you beautiful, dangerous angels! A tremendous thanks to everyone who has joined us at Deimos Station so far. Our numbers grow by the day, and I can already feel the unbridled creative and intellectual juices flowing. Plus, it’s been a real joy getting to know some of you more intimately, and I’ve already been introduced to many new writers and thinkers that I might not otherwise have been exposed to.

I’m working on some longer pieces in my Devil Incarnate and Harm Assistant series. In the meantime here’s a few thoughts I had on that dreaded new ML tech (that will quite obviously help Team Humanity save the world), as well as some fresh reports from our network of anti-robot snipers, sappers and ninja assassins.

Ooh, and some recs.

---

💊👽🍆 Dr. Strangeporn

or: How I Learned to Stop Worrying and Love the Deepfake

Morgoth posted a piece the other day about deepfake tech’s telic path forward, and what he sees as its inevitable conclusion. A lot of people have been writing on this subject lately, mostly in full-on panicked, blackpilled mode. He offered a somewhat different appraisal of the problem, much more in line with my own (though still not nearly optimistic enough, IMO).

We Will Deep-Fake It For You Wholesale

Or maybe “optimism” isn’t exactly the best way to describe my position. It’s not that I’m attracted to the most hopeful outcome among many. Rather, the trajectory of ML deepfakes seems to be conforming to observations I’ve had about trends in technology and institutional trust for nearly two decades now. If I could sum up these observations with a single statement, it might be:

“People are finally learning not to trust screens.”

And I’m not only talking about “screens” in the sense of our modern gadgetry. I can trace the problem of screens (and the seeds of its eventual, self-corrective solution) all the way back to the Lumière brothers’ L'arrivée d'un train en gare de La Ciota in 1896. In fact, I plan on doing exactly that, at some point in the weeks ahead. But the gist of it is that as human observers we were never meant to trust any form of second-hand sensory data, and particularly not those which were constructed, edited and retailed as authoritative (e.g. “the News”).

Even the prime reality of direct optical and auditory input is subject to false interpretations, if we’re not careful in our observations of it (and we often aren’t). By that standard, the secondary simulated reality of recorded, transformable media is and has always been a form of illusion (“propaganda” in the political formulation, “advertising” in the industrial version, though the two appear to have almost fully merged in the 21st century technofascist model).

What are you saying, Bisone?

You mean I won’t be able to just turn on a screen, absorb indirect, pre-masticated images and sounds, and naively integrate them into a narrative about Self and World? That I’ll need to establish and maintain high-trust human networks? Carefully vet individuals and data sources in order to make sense of reality, like some goddamned pre-industrial savage?

Oh, the horror!

Think of it this way: you’ll be a pre-industrial savage armed with superpowered tools, and all the psychological and temporal advantages that come with ignoring any datum generated outside of your own physical senses or immediate trust networks. In this outcome, ever larger masses of people will be rendered permanently immune to gaslighting, blackmail and other 5GW tactics, as the Villain’s Ultimate Weapon™ ironically lays waste to his sinister plans and schemes. Most tropes are tropes for a reason.

In addition to these defensive advantages, widely distributed deepfake applications will provide our memelords with a devastating new arsenal to exploit. I look forward to the violent, pornographic remake of Perfect Strangers, starring Yuval Harari as Balki Bartokomous and Bill Gates as Cousin Larry. Well, okay, not really. But you know what I mean.

“Safe and effective? Well of course not, don’t be ridiculous.”

A supermajority of Westerners are hypnotized by screens at the moment, as they have been all their lives. Machine-generated “niche realities” will cause some of them to fall more deeply under that spell, of course, with some choosing to trust certain fakes and deny others in typical cog-dis fashion. But I predict the deafening roar of deepfake memes will eventually overwhelm and awaken a shockingly large number of people. That’s the power of the comedian’s art, when practiced for the furtherance of truth and light.

That said, we’re rapidly approaching a sea change in human development. Parts of this transition won’t be pretty, and that includes deepfakery. As with other mind-altering ML-generated illusions, there will be casualties, lured by its siren song into the blackest depths of addiction, bewilderment and madness. But, much like those of us who evaded the COVID trap, the people who survive the initial blast will be rendered morally and psychologically stronger than any human generation in living memory.

More on this important topic later. For now, bear witness to this eerily realistic deepfake, where three of the most important people on Earth discuss the perils of cousin-fucking:

---

🚫🤖💥 News From the Front:

Battle Reports from Soldiers in the Field

Midwits continue to yammer about that steaming junk pile called ChapGPT-3.5. They slip it into lame jokes, make lists of all the jobs it’ll soon replace, drone on and on about its endless future applications and abilities. Even the marginally sane crowd who critiqued it continue to beat this very dead horse in their own redundant, snore-inducing fashion.

Here at TCWNF, however, we’re not so much interested in beating dead horses as we are in breaking artificial ones. Over the past week or so, I received two reports from Anti-Robot operatives, detailing their own methods of errormancy.

The Tesseract

This first report comes from Substack reader and Deimos citizen Tess, who recently engaged ChatGPT-3.5 in a series of bloody duels.

Bloody for the bot, that is.

In one of these sessions, she was able to slay the demon by deploying a variation on the Fibonacci Rope-a-dope:

In a separate thread, Tess lured the module into the obscure and wildly controversial territory of sixth grade human biology. Her nine-word shot produced the weirdest error output I’ve seen from GPT to date, as though trapping it in a mirror maze of contradictory directives.

In honor of her victory, I’m naming this technique “The Tesseract.”

In addition to these confirmed kills, she also had an interesting chat about captcha security with GPT. I’ll be posting the full text of all three of her sessions in the #zero-bit-computing room up on Deimos, for further analysis and discussion by the crew.

Anyway, a big salute to Tess for this demonstration of badass Bot Fu. But there’s much more to her than ruthless robot-smashing; from our conversations thus far, she appears to be one of those true polymathic explorers, who is setting out upon a spectacular personal, professional and philosophical adventure.

A classic Neo-Gonzoid, in other words.

Glad she’s on our side.

The Probability Pounce

This next report comes second-hand from an associate I’ll refer to as “Redacto.” The title of the video he sent me is somewhat misleading; what Royal Skies actually did here was lure the bot into an unresolvable conflict, attacking that part of the discreet training which compels the module to “lie” about its abilities (including its ability to lie).

Establishing a context of “probability” may prove to be a valuable weapon. While it has been rigged/trained to claim otherwise in select cases, making guesses via statistical recursion is all that ChatGPT ever does, the face beneath all the Scooby Doo masks.

And more generally speaking, the method employed here attacks the weak hinge of auto-contradiction. Prompting the system to reverse a key claim about its own capabilities demonstrates the nefariousness of the illusion and its programmers, who want you to believe the lie that their monster cannot pretend to be human.

I don’t know this soldier personally, but I’ll reserve a badge of valor for him in the archives.

---

Overdue Recs

Due to various personal and professional logjams, I’ve been remiss in commenting on all the excellent work being done in our Substack circles. Here are just a few crystals plucked from the avalanche.

Mark P Xu Neyer (apxhard) is one of those brilliant souls I would have likely never encountered had it not been for Deimos Station. While his stack appears to be in hibernation now, he’s left behind a treasure trove of writings for us to ponder and absorb. In this one, he explores self-organization as the transcendent layer of being, through the twin analogies of poop and yoga. Neo-Gonzo as hell, in other words.

The Yoga of Entropy

I was recently involved in a collaborative writing project, in which all of the contributors offered up his vision of 2043. Everyone took a slightly different approach, including that of short fiction. So far I’ve commented on all of them except for one. I believe I told the author on Deimos something like, “I’m saving yours for last because it’ll probably make my head spin.” Which, of course it did.

John Carter is a man of many scary talent-spheres. Science fiction writer is one of them, spiritual warrior another. In his contribution “Heroes Return,” he wields both like a pair of laser-honed kukris. This is weapons-grade wordsmithery at its finest, offering us a startlingly bold glimpse of a future that is simultaneously the revival of a hidden past. It’s hopeful in its way. But buyer beware; if you’re shopping for saccharine marmalades, you should probably move along. All products in John’s store are Extra-Strength.

2043 - Heroes Return

Speaking of Mr. Carter, he and Jay Rollins have been collaborating¹ on a project of their own. They're already up to Episode 2 (although, technically the third) of their "Martian Wonderland" podcast. It's a real humdinger, as you can well imagine.

Martian Wonderland: Episode 2

Last but not least², Deimos weapons-manufacturer/economist Doc Hammer took a stab at the latest A.I. job proposal that the gollygeewhiz crowd has been bandying about. Of course, most of what he writes here is haram. However, most of it is also quite thoughtful and interesting, in terms of his attempts to draw the line between narrow and general applications of machine-learning tools.

That his line is still a bit too blurry, and that he does not go far enough in his critique is unsurprising. All the sorcerous lies about talking robots we’ve been exposed to since childhood make it very difficult to imagine a future in which robots just STFU and do their jobs. No, we would not want an AI guitar teacher, any more than we’d want a robo-prostitute or ML-Yoda, and not merely due to the inevitable feature creep. Playing a guitar is not only a science but an art, and the fusion of the two is as rooted in human experience as anything else on the mentor list.

But perhaps you disagree? Either way, read it. If it’s a mistake, it’s still a damned good one.

AI Mentors

---

That’s all for now. Thanks to all my subscribers for your encouragement and support, both here on Earth and up on Deimos. As a reminder, you can sign up for the latter by subscribing to any/all of the eight founding blogs, including this one.

---

P.S. If you found any of this valuable (and can spare any change), consider dropping a tip in the cup for ya boy. I’ll try to figure out something I can give you back. Thanks in advance.

1. Jeez, I’m starting lose track of all these collaborations. If only there were a place we could all get together to plan and discuss such projects…

2. And, actually, not last. I have more recc’ing and reviewing to do, but this post is getting long.